A five-minute review habit for checking AI output before it becomes a message, document, decision, or workflow change.
Before you paste, send, publish, or act on AI output, pause for five checks: source, scope, numbers, voice, and consequence. The habit takes five minutes and catches the most common failure mode in everyday AI work: a draft that sounds finished before it has earned trust.
AI is good at producing work that looks complete. That is useful when you need momentum, but it creates a quiet risk: your brain relaxes because the shape looks right. The paragraph is polished. The bullets are tidy. The recommendation has a confident tone. None of that proves the output is true, complete, appropriate, or ready to use.
The verification pause is a deliberately small habit. It does not turn you into an auditor. It gives you one repeatable way to slow down at the exact moment AI output is most tempting to accept: right after it saves you time.
1. Source. Ask: where did this claim come from? If the answer is a file, link, ticket, transcript, or table, keep the claim. If the answer is "the model probably knows," mark it as unverified or remove it.
2. Scope. Check whether the output stayed inside the job you gave it. AI often widens the frame in helpful-looking ways: adding extra recommendations, inventing next steps, or solving a slightly different problem. Delete anything that was not requested.
3. Numbers. Treat every number as guilty until checked. Dates, counts, totals, percentages, prices, deadlines, and rankings are where small AI errors become expensive. Recalculate or trace them back to the source before they travel further.
4. Voice. Read one paragraph out loud. Does it sound like your team, your customer, your standard, or the person whose name will sit on it? If not, the output may be factually useful but socially wrong.
5. Consequence. Ask what happens if this is wrong. Low-consequence drafts can move fast. Anything that affects money, customers, hiring, compliance, security, legal commitments, or public trust gets a slower review path.
Use this after an AI tool gives you a draft, summary, analysis, or recommendation. Paste the output underneath it.
Review the output below before I use it.
Do not rewrite it yet.
Check it against five categories:
1. Source: which claims need evidence?
2. Scope: what did it add that I did not ask for?
3. Numbers: which dates, counts, totals, percentages, or rankings need verification?
4. Voice: where does the tone not fit the audience?
5. Consequence: which parts need human review before use?
Return a short table with:
- issue
- why it matters
- suggested fix
- risk level: low, medium, or high
Output to review:
[paste the AI output here]Use it before the output leaves your private workspace. Internal notes can be rough. Shared documents, customer replies, leadership summaries, tickets, and project decisions deserve a pause.
Use it when the draft is unusually smooth. Polish can hide weak reasoning. If a document feels too easy to approve, that is exactly when the source and numbers checks earn their keep.
Use it when you are tired. The pause is a guardrail for your attention, not just for the model. It gives your future self a little extra protection at the end of a long day.
Try the pause on one real output today. Do not optimise the prompt yet. Just run the five checks, make the smallest necessary fixes, and notice which category caught the most risk. That is the skill to strengthen next.